Dario Amodei said Anthropic does not oppose open-weight AI models, but wants tighter frontier controls.
Why it matters: The comments clarify Anthropic's position in a major AI policy fight over openness and safety. They also show where frontier labs want regulators to focus - model release practices, testing requirements and chip controls.
- On July 27, Amodei said Anthropic has "never advocated for a ban on open-weights models," per TechCrunch and AP.
- He said open-weight models without dangerous capabilities are "a public good," while frontier systems need stronger safety testing and controls.
- Amodei backed tighter chip export controls, action against "industrial-scale distillation" and mandatory safety testing for sufficiently capable models.
- AP reported that Amodei warned advanced AI could be used by authoritarian governments, including the Chinese Communist Party, for military advantage or repression.
Anthropic CEO Dario Amodei clarified his company's stance on open-weight AI models in a July 27 post, saying Anthropic has "never advocated for a ban on open-weights models", according to TechCrunch. AP separately reported that Amodei said open-weight models can be useful and that Anthropic is not seeking a blanket ban.
He drew a line between models that do not pose major capabilities risks and frontier systems that may need tighter safeguards. In his framing, open-weight models outside that risk tier can be a public good, while more capable systems should face stronger testing and release controls.
Amodei's central concern is national security. AP reported that he warned advanced AI could be used by authoritarian governments, including the Chinese Communist Party, to gain military advantage or deepen repression, and that he also raised biosecurity risks in his remarks.
On policy, Amodei backed tighter chip export controls, restrictions on industrial-scale distillation and mandatory safety testing for sufficiently capable models. In practice, industrial-scale distillation means using a powerful model's outputs to train a cheaper, more accessible copy that can preserve much of the original system's capability.
The broader debate comes as Chinese open-source AI models from firms including Z.ai, Moonshot and Alibaba approach frontier performance, AP reported. The UK AI Security Institute has separately warned that open-weight models can be modified, used without oversight and distributed irreversibly, making them harder to manage than closed systems, in its research at the UK AI Security Institute.
An open letter signed by Nvidia and more than two dozen other companies pushed back on broad, premature restrictions on open-weight AI, according to TechCrunch's coverage.
By the numbers
- July 27 - date Amodei clarified Anthropic's position.
- More than two dozen - companies that signed the open letter pushing back on broad restrictions, per TechCrunch.
Yes, but: The policy fight is not over openness alone - Amodei still argues that some frontier models need tighter controls, testing and export limits.
What's next: No specific deadline was given, but the debate is likely to continue as regulators weigh model release rules, testing requirements and chip export policy.